At the heart of this escalating tension is the experience of clinicians like Yolanda Troublefield, MD, JD, an attending surgeon and otolaryngologist at Southcoast Physicians Group in North Dartmouth, Massachusetts. A member of the American Academy of Otolaryngology–Head and Neck Surgery’s Physician Payment and Policy Workgroup, Dr. Troublefield recently encountered a situation that epitomizes the new reality of prior authorization. She sought approval for an adenoidectomy with ear tube placement for an eight-year-old child suffering from recurrent otitis media. The medical necessity was clear, supported by extensive documentation from both primary care and otolaryngology, including a relevant audiogram. Even the patient’s parents, both physicians themselves, understood the standard of care being recommended.
"We’ve been doing tonsils and adenoids and ventilation tubes for the last 50 years—seems pretty straightforward," Dr. Troublefield remarked, highlighting the routine nature of the procedure. "Every single ‘i’ has been dotted, ‘t’ has been crossed." Yet, despite this thorough preparation, the prior authorization request was denied. Dr. Troublefield attributes this perplexing denial to the burgeoning use of artificial intelligence by insurance companies. She observed that the large language model employed by the insurer appeared to parse only a portion of the medical documentation, failing to grasp the complete clinical picture. In this instance, while the overall diagnosis was "recurrent otitis media," the assessment-and-plan section of the note had also noted "acute otitis media" to indicate the immediate need for antibiotics during that specific visit. This nuanced distinction, easily understood by a human reviewer, was apparently misinterpreted by the AI, leading to the denial.
"What we’re seeing is faster denial of claims because of the use of AI," Dr. Troublefield stated, likening the AI’s settings to a dial that can be adjusted for denial rates. "You can set the dial in different ways. You can set it high, medium, or low. And they’re set high for denials." This scenario necessitates further intervention, potentially a peer-to-peer conversation with a medical director at the insurance company, which disrupts Dr. Troublefield’s schedule and invariably delays essential patient care.
Denials at Machine Speed: The Dawn of Automated Obstruction
The integration of AI into prior authorization processes, and increasingly into retrospective claim reviews or "clawbacks," has led to an unprecedented acceleration of denials. Physicians report that these rejections are often delivered with perplexing logic, compelling them to expend more administrative effort to secure coverage for necessary patient treatments. This trend is leaving healthcare providers increasingly exasperated, as health practices adopt AI for both submitting documentation and responding to denials, creating automated exchanges with potentially far-reaching and unforeseen consequences for the healthcare system.
A recent American Medical Association (AMA) physician survey underscored these growing concerns. A significant 60% of physicians surveyed expressed apprehension that AI is already increasing, or will increase, prior authorization denial rates. Furthermore, 55% of respondents indicated that prior authorization frequently or always delays access to necessary care, while a staggering 79% reported that prior authorization sometimes leads to patients abandoning treatment altogether. Alarmingly, 26% of physicians revealed that prior authorization has resulted in a serious adverse event for a patient under their care, according to AMA data.
Bruce Scott, MD, a former president of the AMA and an otolaryngologist at ENT Care Centers in Louisville, Kentucky, shared his perspective. "I think physicians were very hopeful that AI was going to be a boost, a solution if you will, to reduce the administrative burdens that we all face," he said. However, this optimism has largely been unmet. "Eventually, their bot is going to talk to my bot," Dr. Scott predicted, illustrating the escalating technological engagement. "Because my bot is going to tell me how to document so I get authorization, and then their AI is going to get even smarter and deny that, so I get a denial letter back. And in the meantime, the physicians and patients are stuck in the middle, and that’s the problem."
The timeline of this shift is relatively recent. While AI has been explored in healthcare for years, its aggressive deployment in administrative functions like prior authorization has gained significant momentum in the past two to three years. This acceleration has caught many by surprise, particularly given the initial promise of AI to alleviate administrative burdens rather than exacerbate them.
Regulators Enter the AI Fight: A Patchwork of Oversight
In response to the growing concerns and observed impacts of AI in claim adjudication, a wave of new state laws has emerged, or is slated to take effect, to regulate the use of AI in the denial of claims. These legislative efforts aim to reintroduce a human-centric approach to medical necessity determinations. For instance, an Alabama law mandates that insurers base decisions on a patient’s unique circumstances, prohibiting reliance solely on aggregated group datasets. Similarly, an Indiana law explicitly prohibits insurers from using AI as the exclusive basis for downcoding a claim, a practice that can lead to reduced reimbursement. In Washington state, a new law requires that all claim determinations be made by licensed and qualified health professionals, effectively placing a human safeguard on AI-driven decisions. These are but a few examples of a growing trend toward legislative intervention.
Beyond state-level action, the federal government is also beginning to leverage AI within its own health insurance programs. The Centers for Medicare and Medicaid Services (CMS) has launched the WISeR (Wasteful and Inappropriate Service Reduction) program. This initiative, which commenced in January in six pilot states—New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington—employs AI alongside human review to "reduce clinically unsupported care by working with companies experienced in using enhanced technologies to expedite and improve the review process for a pre-selected set of services that are vulnerable to fraud, waste, and abuse." The stated goal is to streamline the identification and reduction of waste and abuse, though its implementation raises questions about its impact on prior authorization processes for Medicare beneficiaries.
Fighting Automation with Automation: The Provider’s Response
Faced with the increasing automation of denial processes by insurers, many physicians find themselves with little choice but to embrace similar technological solutions. Bradford Bichey, MD, a rhinologist at Indiana Sinus Centers, has taken this approach to an innovative extreme by developing an AI product designed to streamline physician office operations, including the critical task of generating documentation for prior authorization submissions.
Dr. Bichey recounted how, a few years ago, he began to observe a dramatic surge in clawback attempts, where insurers retrospectively questioned and sought repayment for services they deemed improperly covered. What was once a manageable handful of requests per period escalated to dozens, often accompanied by what he described as "early experimentation" with AI in the insurers’ communications. "The verbiage was slightly off," he noted, recalling one particularly "hateful" letter he received, though he acknowledged that these communications have since become more professional in tone.
In response to this challenge, Dr. Bichey developed "Blue," an AI product under his company, Nemedic. This software not only generates a clinical note based on patient encounters but also creates a prior authorization document meticulously tailored to the specific requirements of individual insurers for a given indication. For example, if the term "United" is mentioned during a patient visit, the software automatically cues itself to prepare a document compliant with United Healthcare’s specific protocols, drawing from publicly available insurer documentation. These prior authorization submissions often demand extensive justification and detail far beyond a standard clinical note. For a nasal endoscopy, the AI might generate language such as, "routine anterior examination was insufficient to visualize the deeper nasal passage and assess for ongoing mucosal disease. Diagnostic rigid nasal endoscopy was medically necessary to direct visualization of the sinonasal mucosa," and crucially, includes exclusions like, "this procedure was not performed for routine screening, but to inform management."
"You have to know [the insurer’s] inclusion and exclusion," Dr. Bichey emphasized, highlighting the complexity for human staff. "And it’s almost impossible for someone who does billing to constantly keep up with all these plans because we see hundreds of different types of insurance." He finds that appealing denials, when necessary, is often straightforward with this enhanced documentation, as it typically involves simply pointing to information that was overlooked or ignored in the initial AI review. "When we do this, we just resubmit the same note and say, ‘You’re wrong—look at paragraph three,’" Dr. Bichey explained, noting that this approach has eliminated the need for peer-to-peer calls in his practice for the past two years, thanks to the significantly improved upfront documentation.
Not an AI Problem, but a Governance Problem: The Underlying Dynamic
While the surge in denials might seem like an inherent failing of AI technology, experts suggest the issue lies deeper, within the governance structures surrounding its implementation. Matthew Crowson, MD, assistant professor of otolaryngology–head and neck surgery at Harvard Medical School and director of clinical informatics and artificial intelligence at Massachusetts Eye and Ear, argues that AI itself is not the root cause of the problem. "It’s not so much that AI is the problem," he stated. "It’s more the governance around this stuff, or lack thereof."
Dr. Crowson posits that AI has not resolved the fundamental adversarial dynamic between healthcare providers and payers. Instead, "It’s turning into a battle of the bots," he observed. "The application of AI by providers and payers is cutting both ways. It doesn’t resolve the fundamental tension between approving and denying based on medical necessity. It’s scaling up existing decision-making workflows faster."
However, Dr. Crowson also acknowledges that AI has demonstrated potential for improving administrative efficiency in ways that can benefit patients. For instance, when a prescribed medication is not on an insurance plan’s formulary, a nurse or medical assistant would traditionally face a laborious process of gathering supporting information and completing appeal forms. AI can now draft a contextualized appeal rapidly, potentially expediting medication access for patients. "What would take a human process maybe seven business days to do, now you can do it in one," he said. "I think administratively, it’s massively speeding up our ability to respond quickly."
Efficiency Gains—or Just More Activity? The Unclear Impact
A report released earlier this year by the Peterson Health Technology Institute, based on workshops involving a diverse group of healthcare stakeholders including senior leaders from healthcare systems, health plans, technology developers, investment firms, and federal agencies, presented a nuanced and somewhat uncertain picture of AI’s overall impact on the U.S. medical landscape.
The report indicates that while AI may reduce the cost of obtaining prior authorizations for individual organizations, it has not demonstrably lowered overall system-wide costs. Furthermore, the deployment of AI by providers has been observed to increase "billing intensity," leading to more frequent identification of severe diagnoses and advanced treatments, which in turn can drive up costs. Workshop participants expressed concern that the optimized use of AI by both insurers and providers risks making the entire prior authorization process "more activity-intensive" rather than truly more efficient. One anonymous healthcare provider quoted in the report lamented, "Bots don’t get tired of asking questions, so my review queue keeps growing."
Patients Caught in the Middle: The Human Cost of Automation
Amidst this technological evolution, the prior authorization process continues to strain the vital relationship between physicians and their patients. Dr. Troublefield notes that complex discussions with patients about insurance coverage requirements and the hurdles involved have become a daily fixture, significantly impacting clinic schedules. "It takes more time—I don’t think any of us runs on schedule at all," she stated. "I would say five years ago, we probably could run on schedule. Not anymore. It never, ever, ever happens." This constant delay forces physicians into a position of apology. "We’re finding that we’re constantly having to apologize, apologize, apologize. And that’s not the way people want to be treated."
Dr. Scott recounted a distressing experience with a patient who had a tumor growing in her maxillary sinus. After finally accepting the necessity of surgery, the patient received a denial letter from her insurer stating the procedure was not approved because she had not yet undergone a course of antibiotics—a "preposterous assessment," according to Dr. Scott. The patient, influenced by the insurer’s communication, then questioned Dr. Scott, asking if they should try antibiotics instead. This situation forced Dr. Scott to re-establish trust with his patient, a process he described as challenging and disheartening.
He advocates for more frequent appeals of denied claims, noting that his practice appeals every single denial. However, he acknowledges the reality of physician burnout, which can leave practitioners feeling unable to engage in a constant battle with insurers. Dr. Scott and the AMA are actively pushing for greater physician involvement in the development of regulations and processes surrounding AI in healthcare. "Will it work in my practice? Will I get sued? Will I get paid?" he mused, reflecting the pervasive uncertainty among clinicians.
A Fork in the Road for Prior Authorization: Navigating the Future
Dr. Crowson suggests that the field is approaching a critical juncture, a fork in the road for the future of prior authorization. One path leads to a scenario where both providers and payers effectively utilize AI to process routine paperwork and basic tasks more rapidly, freeing up human resources to focus on more complex and critical cases. The alternative path, he warns, is an "escalating compliance and gaming war," where each side continually refines its AI to gain an advantage over the other.
He expresses hope for a "pragmatic, middle-of-the-road approach," advocating for the use of AI where it excels: automating administrative monotony and other simple tasks that can be safely and efficiently scaled. Dr. Troublefield, while concerned that AI might introduce new requirements for physicians to use specific "buzzwords," finds the concept of software that automatically generates templates based on insurer requirements highly appealing, provided it is practical and user-friendly.
She remains cautiously optimistic that technological advancements, coupled with intelligent legislation—such as robust regulations governing AI’s role in insurance decisions—could offer solutions. She also believes that amplifying patient voices, and even celebrity endorsements, could significantly bolster legislative efforts. Ultimately, however, she concludes that physicians must adapt to these evolving systems. "The system is changing, and either you have to be part of the change, or you’re going to get left behind," Dr. Troublefield stated, drawing a parallel to the advent of the automobile. "It’s the same thing as when the automobile happened, right? You’re not stopping it, so you have to figure out ways within the system to make sure that what you want, and what your goals are, are actually achieved."
